Sloan Management Review Reprint Series EFFECTIVE SUPPLY CHAIN MANAGEMENT Tom Davis ‘Summer 1993 Volume 34 Number 4Effective Supply Chain Management om Davis 1 A-TIME OF SHORTENING PRODUCT LIFE CYCLES, COMPLEX CORPORATE JOINT VEN- "TURES, AND STIFFENING REQUIREMENTS FOR CUSTOMER SERVICE, IT 1S NECESSARY TO + the complete scope of supply chain management, from supplier of raw materi als, through factories and warehouses, to demand in a store for a finished product. Hewlett-Packard has developed a framework for addressing the uncertainty that plagues the performance of suppliers, the reliability of manufacturing and transportation process- 5, and the changing desires of customers. The author describes several cases in which en- tite product families have been reevaluated in a supply chain context. The methodology he presents should help others to manage their own supply chains more successfully. ° Tom Davis isa proces tech- nology manager at Hesolee- Packard, —_— tered phrase these days, as managers strive to im- prove factory performance. The trouble is that all t00 often the real meaning is lost. Instead, a casual ob- server might interpret the activities at the factory as ev dence of an intensive effort to improve supplier manage- Good supplier management, while praiseworthy, docs not constitute good supply chain management without a concurrent effort to manage the rest of the aspects of de- livering products to customers. In this article, I will pre- sent a complete supply chain management methodolo- gy. This approach, developed at Hewlett-Packard, will enable a manufacturing operation wo better manage its supply chain, ultimately improving customer satisfac- tion levels while reducing overall costs. Hewlett-Packard has successfully used this methodolo- gy and is making efforts to implement the practice of good supply chain management at all its divisions, HP identified the need to improve its process for manufactur- S= chain management is a frequently encoun- ing and delivering products to customers as profit mar- gins suffered pressure from increasing competition. Other factors have contributed to a renewed focus, namely: * More instances of multisi manufacturing, where sev- ccral independent entities are involved in the production and delivery process: SLOAN MANAGEMENT REVIEW/SUMMER 1993 + Increasingly cut-throat marketing channels, such as in- dependent computer dealers; + The maturation of the world economy, with heighten- ing demand for “local” products; + Competitive pressures to provide exceptional customer service, including quick, reliable delivery. HP's methodology has led to major changes in the way it does business. In the process, though, we have also discovered some appealing side effects. The proper execution of this methodology leads to improved team- work and cooperation among employees, especially those normally separated by either business furiction ot geography. We've also discovered that this methodology leads to greatly improved customer focus, in addition t0 better relationships with suppliers In the following pages Iwill introduce HPs methodol- ogy. First, I will identify the problems inherent in supply chain management. Then, I will describe some ley insights and briefly describe an analytical tool chat we have devel- ‘oped to support a rational analysis of real supply chains. nally, I will lose with representative case studies describ- g HP successes with “teal” supply chain management. Responding to Uncertainty Large manufacturing companies are “hostage ro complex Davis 35ity,” according to Lew Platt, HP's president and CEO. | ‘The nature of the complexity is evident in a review of material flows for a complicated product. Multiple sup- pliers ship to manufacturing sites with varying regularity. There, subassemblies and final products are made by complicated and somewhat uncertain processes. | Products are then shipped to direct customers, OEM customers, and facturing divis first factory's ‘component materials), ‘The scene is further confused by | eral OEMs" (downstream m: wns of the same company creating the | nal” product as but one of their many neertainty propagates through a | manufacturing network. the wealth of transportation options available: planes, | trains, rucks, and ships, And, of course, multiple carr- | ers convey products t0 customers spread across the globe. The permutations (most of which are actually | used) defy proper management. i “The real problem with such a confusing network is | the uncertainty that plagues it. This uncertainty — ob- served on 2 daily basis as late deliveries, machine break- | downs, order cancellations, and the like — leads to in- | creased inventories, In fact, inventory exists more ot less as simple insurance against uncertainy. In the case of a | Single factory this uncertainty, while bothersome and] costly, is relatively easy to overcome with properly sized inventories of raw materials, work in process (WIP), and finished goods (FGI). Using simple statistical ech- | ques, it’s straightforward 10 figure how much to bold | so that, say, 95 percent of the time a customer will find in stock dhe product he or she wanes, despite the uncer- | tainties in the factory process \ However, the problem is much more complicated when one considers the whole network. Uncertainty | propagates through a manufacturing network, Consider | the following case: A supplier of raw silicon to an inte- grated circuit (IC) manufacturer delivers a bad lot, | Without properly sized safety stocks to buffer the produc tion line against such an event, the IC manufacturer in purer manufacturer. The shortage of parts shuts down the computer line, and shipments 0 computer dealers across. | the country are late. A customer walking into his neigh- bothood computer store discovers the machine he wants | is out of stock. However, the opportunistic salesperson in delays its deliveries to one of its customers, a com | 36 Daws sells him a competing machine available then and there. For want of an on-time silicon delivery (pethaps many months earlier), a sale is lost. Now that’s an extreme example, Nearly every manu- facturer carries some inventory to protect against such situations.’ The real difficulty is knowing how much to hold and where to hol it, for the propagation of intrin- sic uncertainties is mathematically complicated. Because previously there was no clear, analytical way to calculate a proper inventory level (and because there was no clear insight into other processes in the supply chain), firms have traditionally relied on a combination of experience and intuition. This has been their only recourse against the vagaries of their supply chains In the face of increasing pressure to reduce invento- ries and simultaneously improve customer service, the different organizations in a supply chain find themselves working at cross-purposes. This is true when the organ zations exist side by side within the same operation as well as when they are physically remote, independent companies. ‘What often happens is this: One organization will re- duce its inventory to reduce cost. Typically such reduc- tions hurt service. To build up service again, heightened pressure is put on suppliers (which might be the line running just across the aisle) to improve their perfor- mance. If this can be achieved only by increasing the ‘supplier's own inventory, the downstream operation’ in- ventory reductions might be completely canceled out. In many instances, a “broken” supply chain will have sub- stantial stock held ar one point ro enable another node in the supply chain to skate by with minimal stock. ‘Overall, there is roo much stock held in the supply chain system. Reapportioning the stock would reduce the over- all inventory investment and improve end-customer ser- vice levels. Unfortunately, most organizations are de- signed to create winners and losers; working together for _yrtem optimization receives little more than lip service" ‘Only analytical techniques — or darned good luck — can precisely tune a supply chain. In our experience at HP even the best-run supply chains could reduce their overall inventory investment by 25 percent by more clev- crly distributing stock within their systems. More typi- cally, 50 percent reductions are possible.’ With the prop- er tools, understanding and improvement will come from the following three steps: + Benchmarking current performance. Ic is essential that the managers of a particular inventory network know just whar performance is possible, given the existing set of operating, characteristics (order review periods, fore- cast accuracy, etc.). Many of the potential savings come SLOAN MANAGEMENT REvIEW/SUMMER 1993simply from the “cuning” process of adjusting an existing system for better performance. Possible met- rics include inventory investment and order fill rate, among others. * Controlling uncertainty. A com- pany can make great strides by un- derstanding the relative impact of different sources of uncertainty in the system and by then working to reduce (or avoid) the impact they have. As in the computer example, | icis important to measure the indi- rect effect of uncertainty on down- stream or upstream nodes in the supply chain. Most managers con- tending with supply chain issues do not adequately address this. * Planning changes. In addition to adjusting the param ‘eters that describe an existing network, it is possible, with the right tools, co study in advance the benefits (or costs) of sweeping changes to the inventory network. New policies and practices might lead to great reduc- tions in cost and/or improved performance. Without an adequate analysis tool, opportunities for change might be lose for want of a credible argument. Modeling Supply Chains ‘When we first started investigating the problem of modeling Hewlett-Packard’s worldwide inventory net- works, we found ourselves grappling with three funda- mental questions. They stemmed from our interest in studying a difficult academic problem and from our de- sire to provide value to the company by providing ap- propriate solutions to real problems. First, could we de- velop a simple, generic framework to describe the supply chains we faced? Second, could we accurately model the propagation of uncertainty up and down the supply chain? Finally, could we create a modeling ap- proach co support strategic decision makers and nov simply provide yet another ‘ool for solving day-to-day operational problems? ‘As our work progressed, we found that the answer was yes to each of the questions. Whar started as ques- tions evolved into the three key elements of our work. ‘They represent our greatest insights 0 the problem of managing supply chains. We repeatedly use a single, generic model to capture the salient features of each ‘node in a supply chain. We carefully account for the up- stream and downstream effect of uncertainty introduced SLOAN Manacenencr REVIEW/SUMMER 1993 ] Supplier sce ala i | EE ~2onns at any node. And we answer questions posed by strate- gists, not tacticians. Model Simply From an analytical point of view, a supply chain is sim- ply a network of material processing cells with the fol- lowing characteristics: supply, transformation, and de- mand (see Figure 1). This model applies at many level. ‘One can view a factory in this way just as easily as one can view an individual process step within that factory. In both cases, the central manufacturing process — whether it’ called “automobile assembly” or “spot-weld- ing station 1092” — requires raw materials from some supplier external to the process. Those materials are then transformed in some manner that adds value, cre- ating a stock of finished goods (though a JIT process nderstanding the impact of variability in the systém is the goal of modern inventory control theory. might have an empty stockpile). Finally, there is de- mand from external customers for the goods produced, Because of the strategic nature of our work, we avoid living, too deeply into the internal workings of an indi vidual factory. ‘An important aspect of our approach is that we model material handling and distribution functions in the same way as more traditional “manufacturing” pro- cesses, They follow the simple rule we apply: materials Daws 37arrive, value is added, and products exit. In a ware- | house, goods arrive in a ruck or on lift, The distribu- tion team adds value by sorting the deliveries, breaking down pallets as needed, and racking the materials. The istribution operation then fills orders for those Finished goods. That spot-welding station might even be one of the customers. i Indude Uncertainty | “The reason we keep inventory is simple. Inventory is in- surance — protection against life in an uncertain world. “To meet our objectives for customer service, we keep a licele extra material around (in what we call “safety stocks”) 50 that service won't be adversely affected when something in the upstream process goes wrong. Under- standing the impact of variability in the system is the goal of modern inventory control theory. Tn fac, there are three jatinct sources of uncertainty that plague supply chains: suppliers, manufacturing, and customers. To understand fully the impact on cus: tomer service and to be able to improve performance, it | is essential that each of these be measured and ad- | dressed. The systemwide impact of each source of un- certainty must also be understood. Figure 2 illustrates how the three sources of uncertainty in the system com: bine to affect deliveries to customers | ‘The first source of variability that leads to holding safery stocks is supplier performance. This is where naive supply chain considerations start and end. Your supplier quotes a lead time, but — alast — he has a JN Under Forecast. Over Customer Demand hard rime getting you his materials exactly when he says he will. A storm delayed a delivery. A machine broke down. One of his suppliers was late. Any number of things could account for the late delivery, bur in the | end, it remains that sometimes those shipments are late (and, of couuse, sometimes theyre eat) ‘Over time, the vagaries of each supplier's delivery performance can | be tracked. Each supplier's on-time performance, average lateness when late, and degree of inconsistency (as ‘measured by the standard deviation of lateness) are che data to main- tain, These data are immensely valuable, for the characteristics of cach supplier tell you how much extra stock you must hold to keep your line up and running reliably. Supplier performance — and the factory's response, in the form of ‘Supplier Performa 38 Dans Figure2 Uncertainty inthe Supply Chain arly On Time Late nce safety stocks — help determine downstream customer “The second source of variability comes in the manu- facturing process itself. Here, a host of familiar problems can stop the flow of materials off the end of the line. “Machines can break down (just as they can for your sup- plier), another project can tie up a key worker, or a com- puter foul-up can route materials to the wrong place. In measuring process performance as it applies «0 overall supply chain performance, the key metrics are frequency of downtime (for the entire process, not just this or that machine), repair time, and the variation in the repair time. Note that again we're focused on a probability distribution of performance, for that is the nature of uncertainty. Sometimes the line runs for a week without shutting down. Other times, it seems we can't keep it running for more than an hour or ewo without a major shutdown. The reliability of the pro- cess — or its lack of reliability — is one of the three principal determinants of downstream customer service and the inventory investment to achieve that service. Customer demand marks the third and final major source of uncertainty in the supply chain. Depending oon the factory’ location in the supply chain, chis may reflect irregular purchases by a fickle public. Or it may stem from irregular orders from an industrial customer responding to its own up-and-down demand’ In any event, «tue build-to-order factories are relatively rare, and most factories keep some stock, filling orders from inventory as needed. The more variable the orders, the more stock required to reliably meet customer demand. Knowing average demand and the variability of that de- mand makes setting inventory goals more scientific and less seat-of-the-pants. Pow Matecals Si / J me Finished Goods 7 NE IK NK Time Late Early On Time Late Yi Manufacturing Customer Deliveries “SLOAN MANAGEMENT REVIEW/SUMMER 1993Unforcunately, week-to-week demand variability is ‘often not the only problem. Because of long lead times, factories build stock in an attempt to meet forecast de- mand. Hewlett-Packard, for instance, often contracts for raw materials eight months or more before the final product will reach finished goods inventory. Thus, cus- tomer service in August depends on HPs ability co an- ticipace demand the preceding January. The effective variability in this case is greatly compounded from the actual variability we will experience come August. ‘We can measure historical order performance and at ‘tempt to use that information to predict future orders Market research can help too. But such a strategy is often little more than a shot in the dark when introduc ing a new product. With ever-shortening product life cycles, our exposure to this final element of variability in the supply chain occurs more frequently. ‘Act Strategically Customer service starts and ends with the customer. The customer feels the full effect of the complex, interacting uncertainties in che supply chain. In Figure 3, we show what weve coined as the “uncertainty cycle” to capture this effect. The forecast — based on historical customer 3s rolling, Materials are purchased and delivered by suppliers. Then the factory takes its tum, producing the finished goods. Finally, the ealy fore- ‘casts are realized in the form of hard orders, and products au shipped to customers. This cycle repeats itself up and ddown the supply chain. I also repeats over time. Serategic initiatives can lessen che impact of uncer- taimty, First, we can reduce the uncertainties using a standard total quality control approach. We can im- prove forecast accuracy through advanced analytical Figure 3_The Uncertainty Cycle nesearn | EE E + Process design + Product design fT f ‘Capacity = Quality SLOAN Manacenmsr Review/SUMMER 1993, Customer eo + Past performance * Matte research + Analytical techniques * Incentive programs techniques.“ We can investigate more reliable trans- portation modes. We can encourage suppliers to per- form more reliably.’ And we can change product designs to stabilize manufacturing processes. The costs associat- ced with these changes can be balanced against the sav- ings of reduced supply chain inventories or improved Unfortunately, we cannot reduce all uncertainties. But other initiatives can redesign the supply chain to re- duce the impact of the uncertainties. For example, ‘opening a new factory can put production closer to key customers, I'l return t0 this point shortly with some specific examples from HP's experience. ‘A sound supply chain modeling methodology can address long-term problems. Focusing on the uncertain- ties in the system and the way they propagate up and down the chain can lead to new operating policies and objectives. Using tactical tools like transportation mod- cls, scheduling algorithms, or fast MRP programs can fine-tune the performance of the system.* Using a Strategic Modeling Tool Ac HP, we have developed a decision support system that allows us to model supply chains analytically. The tool captures the spirit of the model characteristics de- scribed above, though the mathematical details are not appropriate for this discussion” As we explain to ou ternal HP customers, the tool is of litle value unless one grasps the underlying methodology of supply chain ‘modeling. Supply chain performance can be improved substantially without even switching on a computer. Nonetheless, there are a few points related to the use of the analytical tool that are important. Valid Input ‘As noted earlier, there are many per- formance measures, like supplier late- ness and the associated variability, re- to adequately model a supply chain. Ideally, this dara would be col lected routinely. After all, these are | fundamental measures. However, we have yet to work with an organization that has the data readily available. ‘While organizations typically imagine they can collece the necessary data within just a few weeks, it invari- ably takes six months or longer. Perhaps the single greatest reason for this is that businesses have not taken Davs 39)